1 | /******************************************************************************* |
2 | * Copyright 2020-2022 Intel Corporation |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. |
6 | * You may obtain a copy of the License at |
7 | * |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * |
10 | * Unless required by applicable law or agreed to in writing, software |
11 | * distributed under the License is distributed on an "AS IS" BASIS, |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
13 | * See the License for the specific language governing permissions and |
14 | * limitations under the License. |
15 | *******************************************************************************/ |
16 | |
17 | #ifndef GPU_REF_REDUCTION_HPP |
18 | #define GPU_REF_REDUCTION_HPP |
19 | |
20 | #include "common/c_types_map.hpp" |
21 | #include "common/primitive.hpp" |
22 | #include "common/type_helpers.hpp" |
23 | #include "common/utils.hpp" |
24 | #include "gpu/compute/compute.hpp" |
25 | #include "gpu/gpu_primitive.hpp" |
26 | #include "gpu/gpu_reduction_pd.hpp" |
27 | #include "gpu/gpu_resource.hpp" |
28 | #include "gpu/primitive_conf.hpp" |
29 | |
30 | namespace dnnl { |
31 | namespace impl { |
32 | namespace gpu { |
33 | namespace ocl { |
34 | |
35 | struct ref_reduction_t : public gpu_primitive_t { |
36 | using gpu_primitive_t::gpu_primitive_t; |
37 | struct pd_t : public gpu_reduction_pd_t { |
38 | using gpu_reduction_pd_t::gpu_reduction_pd_t; |
39 | |
40 | DECLARE_COMMON_PD_T("reduction_ref:any" , ref_reduction_t); |
41 | |
42 | status_t init(engine_t *engine) { |
43 | using sm = primitive_attr_t::skip_mask_t; |
44 | const auto attr_skip_mask = sm::post_ops | sm::gpu_attr; |
45 | |
46 | const bool ok = set_default_params() == status::success |
47 | && !memory_desc_ndims_ok(src_md(), dst_md()) |
48 | && attr()->has_default_values(attr_skip_mask) |
49 | && post_ops_with_binary_ok(attr(), dst_md()->data_type, 5) |
50 | && attr_.set_default_formats(dst_md(0)) == status::success; |
51 | if (!ok) return status::unimplemented; |
52 | |
53 | return init_conf(engine); |
54 | } |
55 | |
56 | status_t init_conf(engine_t *engine); |
57 | status_t init_kernel_ctx(compute::kernel_ctx_t &kernel_ctx) const; |
58 | |
59 | reduction_conf_t conf; |
60 | }; |
61 | |
62 | status_t init(engine_t *engine) override { |
63 | compute::kernel_ctx_t kernel_ctx; |
64 | |
65 | status_t status = pd()->init_kernel_ctx(kernel_ctx); |
66 | CHECK(status); |
67 | |
68 | status = create_kernel(engine, &kernel, "ref_reduce" , kernel_ctx); |
69 | CHECK(status); |
70 | |
71 | return status::success; |
72 | } |
73 | |
74 | virtual status_t execute(const exec_ctx_t &ctx) const override { |
75 | return execute_ref(ctx); |
76 | } |
77 | |
78 | private: |
79 | status_t execute_ref(const exec_ctx_t &ctx) const; |
80 | const pd_t *pd() const { return (const pd_t *)primitive_t::pd().get(); } |
81 | |
82 | compute::kernel_t kernel; |
83 | }; |
84 | |
85 | } // namespace ocl |
86 | } // namespace gpu |
87 | } // namespace impl |
88 | } // namespace dnnl |
89 | |
90 | #endif |
91 | |